Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
51 tokens/sec
GPT-4o
60 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
8 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Stable Diffusion Exposed: Gender Bias from Prompt to Image (2312.03027v2)

Published 5 Dec 2023 in cs.CV

Abstract: Several studies have raised awareness about social biases in image generative models, demonstrating their predisposition towards stereotypes and imbalances. This paper contributes to this growing body of research by introducing an evaluation protocol that analyzes the impact of gender indicators at every step of the generation process on Stable Diffusion images. Leveraging insights from prior work, we explore how gender indicators not only affect gender presentation but also the representation of objects and layouts within the generated images. Our findings include the existence of differences in the depiction of objects, such as instruments tailored for specific genders, and shifts in overall layouts. We also reveal that neutral prompts tend to produce images more aligned with masculine prompts than their feminine counterparts. We further explore where bias originates through representational disparities and how it manifests in the images via prompt-image dependencies, and provide recommendations for developers and users to mitigate potential bias in image generation.

PDF HTML Abstract
Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown Bookmark Chat Bubble Oval Streamline Icon: https://streamlinehq.com Chat (Pro)
Definition Search Book Streamline Icon: https://streamlinehq.com
References (75)
  1. Inspecting the geographical representativeness of images from text-to-image models. In ICCV, 2023.
  2. A prompt array keeps the bias away: Debiasing vision-language models with adversarial learning. In AACL-IJNCLP, 2022.
  3. Improving image generation with better captions.
  4. Easily accessible text-to-image generation amplifies demographic stereotypes at large scale. In FAccT, 2023.
  5. Natural language processing with Python: analyzing text with the natural language toolkit. 2009.
  6. Multimodal datasets: misogyny, pornography, and malignant stereotypes. arXiv preprint arXiv:2110.01963, 2021.
  7. Into the LAION’s Den: Investigating hate in multimodal datasets. In NeurlPS Datasets and Benchmarks Track, 2023.
  8. G. Bradski. The OpenCV Library. Dr. Dobb’s Journal of Software Tools, 2000.
  9. Language models are few-shot learners. In NeurlPS, 2020.
  10. Semantics derived automatically from language corpora contain human-like biases. Science, 2017.
  11. Extracting training data from diffusion models. In USENIX Security 23, 2023.
  12. Emerging properties in self-supervised vision transformers. In ICCV, 2021.
  13. Dall-Eval: Probing the reasoning skills and social biases of text-to-image generation models. In ICCV, 2023.
  14. CogView: Mastering text-to-image generation via transformers. In NeurlPS, 2021.
  15. CogView2: Faster and better text-to-image generation via hierarchical transformers. In NeurlPS, 2022.
  16. An image is worth 16x16 words: Transformers for image recognition at scale. In ICLR, 2021.
  17. Diffusion self-guidance for controllable image generation. In NeurlPS, 2023.
  18. Erasing concepts from diffusion models. In ICCV, 2023a.
  19. Unified concept editing in diffusion models. arXiv preprint arXiv:2308.14761, 2023b.
  20. Uncurated image-text datasets: Shedding light on demographic bias. In CVPR, 2023.
  21. GenEval: An object-focused framework for evaluating text-to-image alignment. arXiv preprint arXiv:2310.11513, 2023.
  22. Generative adversarial networks. Communications of the ACM, 2020.
  23. DIG In: Evaluating disparities in image generations with indicators for geographic diversity. arXiv preprint arXiv:2308.06198, 2023.
  24. Deep residual learning for image recognition. In CVPR, 2016.
  25. Prompt-to-Prompt image editing with cross-attention control. In ICLR, 2023.
  26. CLIPScore: A reference-free evaluation metric for image captioning. In EMNLP, 2021.
  27. GANs trained by a two time-scale update rule converge to a local nash equilibrium. In NeurlPS, 2017.
  28. Gender and racial bias in visual question answering datasets. In FAccT, 2022.
  29. Denoising diffusion probabilistic models. In NeurlPS, 2020.
  30. FairFace: Face attribute dataset for balanced race, gender, and age for bias measurement and mitigation. In WACV, 2021.
  31. Bias-to-Text: Debiasing unknown visual biases through language interpretation. arXiv preprint arXiv:2301.11104, 2023.
  32. Segment anything. In ICCV, 2023.
  33. Holistic evaluation of text-to-image models. In NeurlPS Datasets and Benchmarks Track, 2023.
  34. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. In ACL, 2020.
  35. BLIP-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In ICML, 2023.
  36. Word-level explanations for analyzing bias in text-to-image models. arXiv preprint arXiv:2306.05500, 2023.
  37. Microsoft COCO: Common objects in context. In ECCV, 2014.
  38. Grounding DINO: Marrying dino with grounded pre-training for open-set object detection. In arXiv preprint arXiv:2303.05499, 2023.
  39. TF-ICON: Diffusion-based training-free cross-domain image composition. In ICCV, 2023.
  40. Stable bias: Analyzing societal representations in diffusion models. In NeurlPS, 2023.
  41. Multimodal composite association score: Measuring gender bias in generative multimodal models. arXiv preprint arXiv:2304.13855, 2023.
  42. Harvey Mannering. Analysing gender bias in text-to-image models using object detection. arXiv preprint arXiv:2307.08025, 2023.
  43. ClipCap: Clip prefix for image captioning. arXiv preprint arXiv:2111.09734, 2021.
  44. Social biases through the text-to-image generation lens. In AIES, 2023.
  45. Toward verifiable and reproducible human evaluation for text-to-image generation. In CVPR, 2023.
  46. BLEU: a method for automatic evaluation of machine translation. In ACL, 2002.
  47. LD-ZNet: A latent diffusion approach for text-based image segmentation. In ICCV, 2023.
  48. Learning transferable visual models from natural language supervision. In ICML, 2021.
  49. Zero-shot text-to-image generation. In ICML, 2021.
  50. Hierarchical text-conditional image generation with CLIP latents. arXiv preprint arXiv:2204.06125, 2022.
  51. Generative adversarial text to image synthesis. In ICML, 2016.
  52. High-resolution image synthesis with latent diffusion models. In CVPR, 2022.
  53. U-Net: Convolutional networks for biomedical image segmentation. In MICCAI, 2015.
  54. Photorealistic text-to-image diffusion models with deep language understanding. In NeurlPS, 2022.
  55. Improved techniques for training gans. In NeurlPS, 2016.
  56. The bias amplification paradox in text-to-image generation. arXiv preprint arXiv:2308.00755, 2023.
  57. Conceptual Captions: A cleaned, hypernymed, image alt-text dataset for automatic image captioning. In ACL, 2018.
  58. TextCaps: a dataset for image captioning with reading comprehension. In ECCV, 2020.
  59. Diffusion art or digital forgery? investigating data replication in diffusion models. In CVPR, 2023.
  60. What the DAAM: Interpreting stable diffusion using cross attention. In ACL, 2023.
  61. DF-GAN: A simple and effective baseline for text-to-image synthesis. In CVPR, 2022.
  62. Stereotypes and smut: The (mis) representation of non-cisgender identities by text-to-image models. In ACL, 2023.
  63. CIDEr: Consensus-based image description evaluation. In CVPR, 2015.
  64. T2IAT: Measuring valence and stereotypical biases in text-to-image generation. In ACL, 2023a.
  65. Evaluating data attribution for text-to-image models. In ICCV, 2023b.
  66. Image quality assessment: from error visibility to structural similarity. IEEE TIP, 2004.
  67. Contrastive language-vision ai models pretrained on web-scraped multimodal data exhibit sexual objectification bias. In FAccT, 2023.
  68. Diffumask: Synthesizing images with pixel-level annotations for semantic segmentation using diffusion models. In ICCV, 2023.
  69. Attention as annotation: Generating images and pseudo-masks for weakly supervised semantic segmentation with diffusion. arXiv preprint arXiv:2309.01369, 2023.
  70. From image descriptions to visual denotations: New similarity metrics for semantic inference over event descriptions. In ACL, 2014.
  71. Scaling autoregressive models for content-rich text-to-image generation. TMLR, 2022.
  72. ITI-GEN: Inclusive text-to-image generation. In CVPR, 2023a.
  73. Recognize anything: A strong image tagging model. arXiv preprint arXiv:2306.03514, 2023b.
  74. Auditing gender presentation differences in text-to-image models. arXiv preprint arXiv:2302.03675, 2023c.
  75. Men also like shopping: Reducing gender bias amplification using corpus-level constraints. In EMNLP, 2017.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (3)
  1. Yankun Wu (4 papers)
  2. Yuta Nakashima (67 papers)
  3. Noa Garcia (33 papers)
Citations (10)
View on Semantic Scholar